Upload config.json with huggingface_hub
Browse files- config.json +54 -0
config.json
ADDED
|
@@ -0,0 +1,54 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"_attn_implementation_autoset": true,
|
| 3 |
+
"_name_or_path": "Alibaba-DAMO-Academy/PixelRefer-7B",
|
| 4 |
+
"architectures": [
|
| 5 |
+
"PixelreferQwen2ForCausalLM"
|
| 6 |
+
],
|
| 7 |
+
"attention_dropout": 0.0,
|
| 8 |
+
"bos_token_id": 151643,
|
| 9 |
+
"eos_token_id": 151645,
|
| 10 |
+
"hidden_act": "silu",
|
| 11 |
+
"hidden_size": 3584,
|
| 12 |
+
"image_aspect_ratio": "qwen2vl",
|
| 13 |
+
"image_size": 28,
|
| 14 |
+
"image_token_index": 151665,
|
| 15 |
+
"image_token_length": 1,
|
| 16 |
+
"initializer_range": 0.02,
|
| 17 |
+
"intermediate_size": 18944,
|
| 18 |
+
"is_alignment": false,
|
| 19 |
+
"llm_lr": 2e-05,
|
| 20 |
+
"mask_num": 32,
|
| 21 |
+
"mask_text_to_visual": true,
|
| 22 |
+
"max_frames": 16,
|
| 23 |
+
"max_position_embeddings": 32768,
|
| 24 |
+
"max_window_layers": 28,
|
| 25 |
+
"mm_hidden_size": 1152,
|
| 26 |
+
"mm_max_length": 2048,
|
| 27 |
+
"mm_projector_lr": 0.0,
|
| 28 |
+
"mm_projector_type": "mlp2x_gelu",
|
| 29 |
+
"mm_vision_encoder": "DAMO-NLP-SG/VL3-SigLIP-NaViT",
|
| 30 |
+
"mm_vision_select_feature": "patch",
|
| 31 |
+
"mm_vision_select_layer": -2,
|
| 32 |
+
"model_type": "pixelrefer_qwen2",
|
| 33 |
+
"num_attention_heads": 28,
|
| 34 |
+
"num_hidden_layers": 28,
|
| 35 |
+
"num_key_value_heads": 4,
|
| 36 |
+
"region_encoder_lr": 0.0,
|
| 37 |
+
"region_token_index": 151668,
|
| 38 |
+
"rms_norm_eps": 1e-06,
|
| 39 |
+
"rope_scaling": null,
|
| 40 |
+
"rope_theta": 1000000.0,
|
| 41 |
+
"sliding_window": null,
|
| 42 |
+
"spatial_merge_size": 2,
|
| 43 |
+
"tie_word_embeddings": false,
|
| 44 |
+
"tokenizer_model_max_length": 8192,
|
| 45 |
+
"tokenizer_padding_side": "right",
|
| 46 |
+
"torch_dtype": "bfloat16",
|
| 47 |
+
"transformers_version": "4.46.3",
|
| 48 |
+
"use_cache": true,
|
| 49 |
+
"use_mm_proj": true,
|
| 50 |
+
"use_sliding_window": false,
|
| 51 |
+
"use_token_compression": true,
|
| 52 |
+
"vision_encoder_lr": 0.0,
|
| 53 |
+
"vocab_size": 151669
|
| 54 |
+
}
|